'The compute bottleneck is massively under appreciated' says Google AI Studio lead: 'I would guess the gap between supply and demand is growing [by a] single digit % every day'

As the AI boom continues to, well, boom, it's not surprising that some are questioning the viability of rampant AI adoption, and the pace at which the hardware supply chain underneath it has to keep up.

One such commenter on the subject is none other than the product lead for Google AI Studio, Logan Kilpatrick. Posting on X, Kilpatrick said: "The compute bottleneck is massively under appreciated."

"I would guess the gap between supply and demand is growing single digit % every day", Kilpatrick continued. "In practice, this will be the rate limit on the impact AI will have in the economy/society."

Certainly, it's not the first time we've heard Google's name in association with massive, seemingly unsustainable AI compute demands. A leaked internal slide shown during a company meeting last year informed one of its teams that AI compute "must double every six months... the next 1000x in 4-5 years."

Simply put, there's only so much compute to go around, and there's a limit to how quickly it can be realistically expanded, even for Google.

TSMC, the manufacturer behind 90-95% of the world's most advanced chips, already looks to be near maximum capacity, resulting in Nvidia CEO Jensen Huang giving the company the hurry up in public.

More facilities look to be on their way—but the idea of it catching up with current demand any time soon seems unlikely. And let's not even mention the DRAM supply crisis, and what that's doing to electronics pricing right now.

So, without major efficiency gains on the software side of things, and with more and more AI tools, agents, and doohickies demanding vast amounts of computing capacity to operate, there seems to be an obvious bottleneck as to how fast the supply chain can keep up. Even if you do end up building data centers in tents.

X user @zazmic has many of the same thoughts: "Honestly, everyone talks about AI changing the world, but if even Google is running into compute limits, that tells you something... you can have all the ideas in the world, but if there aren’t enough chips to run them, progress slows down whether you like it or not.

"AI isn’t just a software story, it’s a supply chain story now, and that simply doesn’t scale overnight."

However, a built-in limiting factor to the mass adoption of compute-heavy AI agents might not be the worst thing.

A report released by Citrini Research earlier this week, which posited a doomsday-like scenario in which the widespread use of autonomous AI agents caused the US economy to falter (and eventually leads to mass protests on the streets), seems to have shaken US market confidence in the tech to a surprising degree.

Speaking personally, I can say that the idea of a built-in limit on how fast AI can expand is somewhat cheering. While the current demand is ruining the availability of PC gaming hardware (and certainly won't be helped by further supply/demand woes), perhaps a built-in limit means I'll be able to keep writing articles for a while yet, yes?

Small wins, folks. I have to take the small, intensely personal wins.